What problem does it solve? When a project opts into multi-model auditing, a single reviewer's verdict may miss defects that another model would catch. This Skill defines how plan-auditor and sync-auditor agents fan a code review out across the codex and GLM backends in parallel, converge their verdicts with the in-session Claude verdict, and fold the result into the audit report without corrupting the independence of each backend's opinion. ## Core Features & Use Cases - Single tool surface: Documents the mcp__moai__audit_multi MCP tool, its input parameters (claude_verdict, target, focus, gates, session_id), and its ConvergenceResult output shape including per-backend verdicts, overall verdict, disagreement flag, and residual risk note. - Independence enforcement: Explains the load-bearing rule that secondary backends receive only the target and focus area, never Claude's full analysis, so codex and GLM remain uncorrelated super-reviews. - Convergence policy: Defines the 4-case table mapping required and advisory gate outcomes to overall pass/fail, plus fail-open behavior when optional backends are missing or erroring. - Use Case: An auditor agent reviewing uncommitted changes in a project configured with audit_model: multi loads this Skill, calls audit_multi with its synthesized Claude verdict, and reports a conservative FAIL naming the failing required backend when the models disagree. ## Quick Start Run a multi-model audit of the uncommitted changes by calling the audit_multi tool with my Claude verdict and fold the convergence result into the audit report.
